Three-Phase Grid-Connected NPC Inverter Based on a Robust Artificial Neural Network Controller

This paper proposes a robust Artificial Neural Network (ANN) controller based on Multilayer Perceptron (MLP) topology to control three-phase Neutral Point Clamped (NPC) inverter for grid-connected application. The proposed MLP Controller (MLPC) is trained by Gradient Descent optimization (GDO) algorithm based on primary data which is obtained from a Sliding Mode Control (SMC) loop. Using the primary data, robust performance of the ANN-based control loop is attained while the control loop is much simpler for real time implementation than SMC. The simulation results achieved by MATLAB/Simulink software prove optimal performance of the proposed MLPC even under unstable conditions which are not existing during training process.